Stacked ensemble model for analyzing mental health disorder from social media data

Mental health issues are detrimental to quality of life and frequently increase the dangers of mental disease like suicidal thoughts, depression, and many are frequently misdiagnosed and remain untreated. Accurate diagnosis of indicators of mental health disorders is especially crucial because they may be life-threatening when left untreated. Therefore, it is essential to have a thorough understanding of the sources to recognize a person's mental illness. Social media is the major source of identifying individual mental diseases, and hence their needs a sophisticated analysis from several angles to prevent death. On the other hand, researchers are examining whether computer methods could track the communication on social networks, which might help in the early identification of mental health issues. This paper creates a new system for analyzing mental health disorders via social media data that avoids the problem from becoming serious. Preprocessing, feature extraction, and classification are the three steps of this approach. First, stemming and stop word removal will be processed as the preprocessing step of input text. From the preprocessed text, Improved Semantic Similarity features, content features, BoVW, features like n-grams, LIWC features, and are extracted. The ensemble classifier, which includes the classifiers from the Bidirectional Gated Recurrent Unit (Bi-GRU), Deep Maxout, and Improved Convolution Neural Network (ICNN), performs a categorization using the extracted characteristics. Also, to improve the efficiency of classification, a training model is introduced in the ICNN termed as self-adaptive Shuffled shepherd optimization method (SASSOA) that tunes the optimal weights. Finally, the efficacy of the projected method is verified to the convolutional procedures.